Advanced Neural Networks for Electromagnetic Modeling and Design
Bing‐Zhong Wang, Li‐Ye Xiao, Wei Shao · 2023
The artificial neural network (ANN), a powerful machine learning method, is a good alternative to full-wave simulations to significantly speed up the optimization design of microwave components, circuits, and antennas. This chapter discusses some advanced techniques suitable for efficient ANN modeling in electromagnetics. First, two semi-supervised learning schemes are introduced to model microwave passive components, based on the dynamic adjustment kernel extreme learning machine and radial basis function neural network. Second, some effective ANN techniques are proposed for antennas and arrays, including multi-parameter modeling in antenna modeling, inverse ANN for multi-objective design, and prior knowledge for periodic array modeling. Third, an autoencoder neural network, which is used to reduce the dimensionality of data from time-domain full-wave simulations, calculates wave propagation in uncertain media.